The Agent Skills Directory

[SAFE]: The skill includes a local Python script scripts/rubric-scorecard.py for calculating evaluation scores. The script safely processes JSON data using standard libraries (json, pathlib) and performs only arithmetic operations without any network activity or command execution.- [SAFE]: The skill defines a workflow for processing external artifacts, which is a common surface for indirect prompt injection. However, the workflow incorporates specific safety measures, including 'Self-Verification Phase-Gate Questions' and a requirement for evidence-based criteria, to ensure the agent remains focused on the rubric rather than being influenced by the content of the artifact. Ingestion occurs via artifact collection in the SKILL.md core loop, and sanitization is handled through safe JSON parsing of the resulting scores.- [SAFE]: Metadata and portability instructions are well-defined and consistent with the skill's purpose. The mentions of platform-specific integration scripts (such as export-gemini-skill.py) in the documentation are for development workflow management and do not involve the automated execution of untrusted code.

agentic-eval